Automatically reduce batch size for small validation/test splits

Author: antihutka

eval.lua

@@ -53,6 +53,7 @@ local loss = 0
 for i = 1, num do
   print(string.format('%s batch %d / %d', opt.split, i, num))
   local x, y = loader:nextBatch(opt.split)
+  N = x:size(1)
   x = x:type(dtype)
   y = y:type(dtype):view(N * T)
   local scores = model:forward(x):view(N * T, -1)

train.lua

@@ -205,9 +205,10 @@ for i = start_i + 1, num_iterations do
     local val_loss = 0
     for j = 1, num_val do
       local xv, yv = loader:nextBatch('val')
+      local N_v = xv:size(1)
       xv = xv:type(dtype)
-      yv = yv:type(dtype):view(N * T)
-      local scores = model:forward(xv):view(N * T, -1)
+      yv = yv:type(dtype):view(N_v * T)
+      local scores = model:forward(xv):view(N_v * T, -1)
       val_loss = val_loss + crit:forward(scores, yv)
     end
     val_loss = val_loss / num_val

util/DataLoader.lua

@@ -24,16 +24,21 @@ function DataLoader:__init(kwargs)
   self.split_sizes = {}
   for split, v in pairs(splits) do
     local num = v:nElement()
-    local extra = num % (N * T)
+    local N_cur = N
+    if (N * T > num - 1) then
+      N_cur = math.floor((num - 1) / T)
+      print(string.format("Not enough %s data, reducing batch size to %d", split, N_cur))
+    end
+    local extra = num % (N_cur * T)
 
     -- Ensure that `vy` is non-empty
     if extra == 0 then
-      extra = N * T
+      extra = N_cur * T
     end
 
     -- Chop out the extra bits at the end to make it evenly divide
-    local vx = v[{{1, num - extra}}]:view(N, -1, T):transpose(1, 2):clone()
-    local vy = v[{{2, num - extra + 1}}]:view(N, -1, T):transpose(1, 2):clone()
+    local vx = v[{{1, num - extra}}]:view(N_cur, -1, T):transpose(1, 2):clone()
+    local vy = v[{{2, num - extra + 1}}]:view(N_cur, -1, T):transpose(1, 2):clone()
 
     self.x_splits[split] = vx
     self.y_splits[split] = vy